CN102855759B - Automatic collecting method of high-resolution satellite remote sensing traffic flow information - Google Patents

Automatic collecting method of high-resolution satellite remote sensing traffic flow information Download PDF

Info

Publication number
CN102855759B
CN102855759B CN 201210344373 CN201210344373A CN102855759B CN 102855759 B CN102855759 B CN 102855759B CN 201210344373 CN201210344373 CN 201210344373 CN 201210344373 A CN201210344373 A CN 201210344373A CN 102855759 B CN102855759 B CN 102855759B
Authority
CN
China
Prior art keywords
vehicle
image
road
traffic flow
traffic
Prior art date
Application number
CN 201210344373
Other languages
Chinese (zh)
Other versions
CN102855759A (en
Inventor
刘亚岚
刘珠妹
任玉环
Original Assignee
中国科学院遥感应用研究所
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority to CN201210232694 priority Critical
Priority to CN201210232694.2 priority
Application filed by 中国科学院遥感应用研究所 filed Critical 中国科学院遥感应用研究所
Priority to CN 201210344373 priority patent/CN102855759B/en
Publication of CN102855759A publication Critical patent/CN102855759A/en
Application granted granted Critical
Publication of CN102855759B publication Critical patent/CN102855759B/en

Links

Abstract

本发明公开了一种高分辨率卫星遥感交通流信息自动采集方法,其步骤:A、预处理,包括全色影像与矢量路网及全色与多光谱影像的配准,道路区域分割与双边滤波增强。 The present invention discloses a high-resolution satellite remote sensing automatic traffic flow information acquisition method, comprising the steps: A, pretreatment, and vector comprising a panchromatic road network and panchromatic and multi-spectral image registration, segmentation and bilateral road filtering enhancements. B、据对步骤A得到的道路区域影像目视判读获得车辆样本特征值,建立车辆遥感影像特征库。 B, according to the sample feature value is obtained for a road vehicle image region obtained in Step A visual interpretation, to establish vehicle remote sensing image feature database. C、对步骤B所得全色影像进行神经网络车辆粗提取与面向对象车辆精提取。 C, step B the resulting full color image of the vehicle for the neural network of the crude extract and extraction of object-oriented fine vehicle. D、利用图像频率域相关匹配法,在多光谱影像中搜索对应步骤C精提取的车辆位置并进行匹配。 D, using a frequency domain correlation image matching method, the step of searching a position corresponding to the vehicle C in the extract derived multispectral image and match. E、据步骤C和步骤D得到的对应车辆位置,计算全色与多光谱影像中同一车辆的位移量,进而估算各交通流参数信息。 E, corresponding to the vehicle according to Step C and Step D to give the position, calculates the amount of displacement of the panchromatic and multi-spectral image of the same vehicle, each of the traffic flow parameters are estimated from information. F、通过精度评价加以验证。 F, verified by the evaluation accuracy. 该方法能实现大范围系列静态和动态交通流信息自动化快速采集,效率更高,简单易行。 This method can achieve a wide range of static and dynamic series of automated rapid traffic flow information collection, more efficient and simple.

Description

高分辨率卫星遥感交通流信息自动采集方法 High-resolution satellite remote sensing automatic traffic flow information collection method

技术领域 FIELD

[0001] 本发明涉及卫星遥感在智能交通领域中的图像处理与分析技术,具体涉及一种高分辨率卫星遥感交通流信息自动采集方法,适应于对大范围路网的交通流监测,快速、自动获取系列静态和动态的交通流信息,可满足大范围宏观交通调查的需要,为城市与区域交通管理与交通规划提供服务。 [0001] The present invention relates to satellite remote sensing image processing and analysis techniques in the field of intelligent transportation, particularly to a high-resolution satellite remote sensing automatic traffic flow information acquisition method, adapted to a wide range of monitoring traffic flow of the road network, fast, automatically obtain series of static and dynamic traffic flow information, to meet the needs of a wide range of macro traffic survey, to provide services to the urban and regional transport planning and traffic management.

背景技术 Background technique

[0002] 近几年来,私人轿车保有量飞速增长,我国越来越多的城市开始面临交通拥堵问题,不仅影响到公众出行、消耗大量的能源,造成环境污染,还影响到城市功能正常发挥和可持续发展。 [0002] In recent years, the rapid growth of private car ownership, more and more Chinese cities began to face traffic congestion problems, not only affect the public, consuming large amounts of energy, causing environmental pollution, but also affects the normal functioning of the city and sustainable development. 智能交通系统(ITS, Intelligent Transportation System)作为解决交通运输安全、效率和拥挤问题的有效工具应运而生。 Intelligent Transportation Systems (ITS, Intelligent Transportation System) emerged as an effective tool to address transportation safety, efficiency and congestion problems. ITS包括交通信息采集、分析与发布三个阶段,其中交通流参数的采集是智能交通系统实施好坏与否的关键。 ITS include traffic information collection, analysis and publication of three stages, collect traffic flow parameters is the key to the implementation of intelligent transportation system is good or bad or not. 交通流参数自动采集方式包括感应线圈、压电式检测器、视频检测器、微波检测器,以及基于有人机、无人机、飞艇等航空遥感与航天卫星遥感的检测,以及基于浮动车的动态采集方式。 Automatic collection of traffic flow parameters embodiment comprises detecting an induction coil, a piezoelectric detector, a video detector, a microwave detector, and based on manned, unmanned aerial vehicles, airships and other aviation and aerospace remote sensing satellite remote sensing, and the dynamic floating cars based acquisition mode.

[0003] 传统的交通流参数采集设备(如电感线圈、视频检测器等)由于成本高、安装复杂、不易移动和维修,目前只在大中型城市监控主干道的交通信息,不适合大范围交通数据的采集。 [0003] Traditional traffic flow parameter acquisition device (e.g., inductor, a video detector, etc.) due to the high cost, complexity of installation, maintenance and easy to move, only the traffic information is currently monitoring medium-sized cities main road, not suitable for large scale traffic data collection. 近几年发展起来且应用较多的浮动车采集技术,一般是将GPS安装在出租车上,要求车辆数量多,成本高,且由于车辆行经路线随机性大、不能保证路网全覆盖,所得到的数据只在一定情况下具有代表性。 Developed in recent years and the application of more floating car collection technology, GPS is generally installed in the taxi, the number of vehicles required, high cost, and because of the randomness of large vehicles passing through the route, the road network can not guarantee full coverage of the the resulting data representative only under certain circumstances. 在实际应用中,通常只对主干道路上的数据进行采集,不能主动获取数据,且数据处理过程复杂。 In practice, usually only the arterial road data acquisition, data acquisition is not active, and the data processing process is complicated. 遥感技术具有宏观、快速、经济、覆盖范围大等独特的数据获取优势,并且由于传感器远离路面,与地面调查相比,既不会影响交通,也不会给地面调查人员带来危险,可大幅度降低实地调查的劳动强度和调查成本,提高交通流参数采集的广度和全局性,因而它在交通领域的应用得到国内外的广泛关注。 Remote sensing technology has a macro, fast, economical, covering a large range of data acquisition and other unique advantages, and because the sensor away from the road, compared with ground surveys, will not affect the traffic, do not give ground Investigators at risk, can be as large reduce labor intensity and magnitude of the survey fieldwork costs, improve traffic flow parameters capture the breadth and global, and therefore its use in the transport sector received extensive attention at home and abroad. 卫星遥感与航空遥感手段相比,数据获取更加经济,数据处理流程更加成熟。 Compared with the satellite remote sensing airborne remote sensing instruments, data acquisition more economical, more mature data processing flow. 因此,利用卫星遥感影像进行交通流信息采集具有较大的潜力与应用前景。 Therefore, the use of satellite remote sensing images of traffic flow information collection has great potential and prospects. 尤其各种高分辨率遥感数据的不断涌现,为实现卫星遥感交通流信息的自动采集带来了更多的机遇。 In particular, a variety of high-resolution remote sensing data continue to emerge, in order to achieve the automatic collection of satellite remote sensing traffic flow information has brought more opportunities. 随着我国高分对地观测计划的实施,将有效缓解对国外高分辨率卫星遥感数据获取的依赖程度,为基于我国自主高分辨率卫星的交通流信息采集创造了良好的条件。 With the implementation of Earth observation program high marks in our country, will effectively ease dependence on foreign high-resolution satellite remote sensing data acquisition, creating good conditions for the traffic flow based on independent information gathering high-resolution satellite in China.

[0004] 目前遥感交通监测需要解决车辆识别和运动车辆在不同遥感图像上位置匹配两个主要问题。 [0004] need to address the current remote sensing traffic monitoring vehicle identification and location of a moving vehicle on a different remote sensing image matching two main issues. 利用高分辨率卫星遥感技术进行交通监测管理方面的研究还不多见,且大多为车辆目标的识别,识别精度和自动化程度均不高;对车辆目标的匹配方法,虽有研究涉及到,但是采用人工目视方法来实现的。 Research management traffic monitoring using high-resolution satellite remote sensing technology also rare, and most of the vehicle target recognition, recognition accuracy and automation are not high; matching vehicle targets, although the study involved, but artificial visual means to achieve. 因此,如何充分利用高分辨率卫星遥感技术的优势,提高现有的车辆检测方法的精度和自动化程度,实现车辆目标的自动匹配,形成一套系统的交通流信息的自动化采集技术方法,以满足交通管理与规划对交通参数采集的实际需求,具有非常重要的实用价值,尤其是对促进我国智能交通系统技术的发展和城市交通管理与规划的水平的提升具有重要的意义。 Thus, how to take advantage of high-resolution satellite remote sensing technology, to improve the accuracy and automation of existing methods of detecting the vehicle, the target vehicle automatic matching, automatic forming method of traffic flow information collection technology of a system to meet traffic management and planning of the actual needs of traffic parameters collection, has a very important practical value, in particular, is of great significance to promote the upgrading of the level of development and urban planning and traffic management and intelligent transportation systems technology in China. [0005] 与本发明直接相关的技术方案: [0005] Technical Solution directly related to the present invention:

[0006] 本发明中涉及车辆检测和车辆匹配两个方面的关键技术。 [0006] The present invention relates to vehicle detection and vehicle match of two key technologies. 与此相关的技术方案分别说明如下。 Related to this aspect are described below.

[0007] 1.车辆检测方面 [0007] 1. The vehicle detection aspects

[0008] “利用遥感影像探测和计数城市道路车辆的方法”(专利公开号:101404119)与本发明的其中一个步骤相关。 [0008] "Remote sensing and image detection urban road vehicle count method" (Patent Publication No: 101404119) a step in which the present invention is related.

[0009] 利用遥感采集交通流信息,首先需得到道路中运行车辆个体信息。 [0009] using remote sensing collect traffic flow information, you first need to obtain information on individual vehicles running on the road. 谭衢霖等人提出了一种利用遥感影像探测和计数城市道路车辆的方法,其主要步骤如下: TAN Qu-lin, who proposed a method for using remote sensing image detection and enumeration of urban road vehicles, the main steps are as follows:

[0010] 第一步,基于道路中心线生成城区道路掩膜,限制车辆探测在道路区域进行; [0010] The first step, based on the road centerline mask generation urban roads, limitations of the vehicle in a road detection region;

[0011] 第二步,对在步骤一中生成的城区道路掩膜的影像进行二次不同尺度的影像分割获得道路车道条带目标层和车辆探测对象基本层。 [0011] The second step, the video urban road mask generated in Step a is carried out at different scales of the image segmentation obtained secondary road lane strip target layer and the base layer of the vehicle to detect objects. 其中,为获得道路车道条带目标层所使用的分割尺寸大于为获得车辆探测对象基本层所使用的分割尺寸; Wherein, in order to obtain a road lane striping size greater than the target layer is used to obtain a target vehicle detection division size of the base layer is used;

[0012] 第三步,在车辆探测对象基本层构建面向对象的模糊分类器对该层的对象进行车辆对象和非车辆对象分类; [0012] The third step, in the vehicle detection target base layer building object-oriented fuzzy classifier vehicle and non-vehicle objects subject layer to the object classification;

[0013] 第四步,在被分类的车辆探测对象基本层上,融合相邻的同类对象生成车辆探测融合对象层,在该层上分类车辆,最终获得完成车辆探测的影像。 [0013] The fourth step in the classified vehicle detection target base layer, generating a fusion of the adjacent vehicle to detect objects of the same fusion target layer, on the layer classification of the vehicle, the vehicle finally obtained complete image detection.

[0014] 本发明与之相比,有以下三点: Comparison [0014] with the present invention, the following three points:

[0015] 1.不仅实用于大范围、快速、自动化采集,而且采集的交通流信息更广泛,不仅能获得车辆数量、类型、分布等静态参数,还可以在车辆匹配的基础上获得交通流速度、流量等其他系列动态交通流参数。 [0015] 1. only practical for large-scale, rapid, automated collection, and traffic flow information collection more widely, not only to obtain static parameters number of vehicles, type, distribution, etc., can also be obtained on the basis of traffic flow speed of the vehicle match other series of dynamic traffic flow and other parameters.

[0016] 2.将神经网络与面向对象分类方法相结合,实现了高分辨率遥感影像的自动化分害I],避免了人工处理需要反复选择分割尺度的局限。 [0016] 2. The object-oriented neural network classification method combining automated Resolution Image partial damage the I], avoiding the need for manual handling of repeated selection segmentation scale limitations.

[0017] 3.引入了车辆遥感影像特征库,实现了车辆目标分类器的阈值自动化选择,从而避免了人工干预阈值选择的主观性和阈值反复调整的复杂性。 [0017] 3. The introduction of the vehicle remote sensing image feature library, to achieve automatic threshold selection vehicle Classifier, thereby avoiding the complexity of manual intervention and subjective selected threshold value is repeatedly adjusted threshold.

[0018] 以上三点优势,有利于提高卫星遥感交通流信息的自动化采集能力,使其更适合大范围路网、更全面的交通流信息的采集。 [0018] The above three points advantage, help improve the ability to automate collection of satellite remote sensing traffic flow information, making it more suitable for large-scale road network and more comprehensive traffic flow information collection.

[0019] 2.车辆匹配方面: [0019] The vehicle in matching:

[0020] 为了利用高分辨率卫星遥感影像进行包括车速等在内的诸多交通流信息(如交通流量、车头距离等)的采集,需要从同一遥感平台获取的具有拍摄时间差的全色影像和多光谱影像中得到车辆目标的位移信息。 [0020] For many traffic flow information (e.g., traffic flow, etc. from the front) of the collection, including a vehicle speed, etc., require high resolution panchromatic satellite remote sensing image from the photographing time difference having the same multi-sensing platform and acquired displacement information obtained spectral image of the target vehicle. 由于同一遥感平台中多光谱影像空间分辨率较全色影像的空间分辨率低,仅依靠影像分类的方法很难精确提取多光谱影像中的车辆目标信息。 Since the same platform, multi-spectral remote sensing image spatial resolution than the low spatial resolution panchromatic image, only to rely on image classification method is difficult to accurately extract the vehicle target information multispectral image. 因此,本发明提出采用图像频率域相关匹配法进行车辆匹配,即根据由全色影像分类得到的已知车辆目标区域来搜索多光谱影像中同一车辆的位置,从而实现两种遥感影像中相同车辆目标的匹配。 Accordingly, the present invention provides a vehicle to match the frequency domain using the image matching correlation method, i.e., to search for a position in the same multi-spectral image obtained by a vehicle according to the classification known to a full-color image the target area of ​​the vehicle, in order to achieve two kinds of remote sensing images in the same vehicle matching goals.

[0021]目前,传统的图像匹配方法是以互相关函数作为匹配准则,以相关系数的大小评价影像匹配位置与模板的相似度程度,从而确定最大相关位置,即最佳匹配位置的过程,即: [0021] Currently, the conventional image matching method is based on a cross correlation function as the matching criterion, the degree of similarity to the template image matching the size and position of the evaluation of the correlation coefficient, to determine the location of maximum correlation, i.e. the best match location process, i.e., :

Figure CN102855759BD00071

[0023] 该方法计算复杂,时间效率较低。 [0023] The method of calculating the complex, time efficiency is low.

发明内容 SUMMARY

[0024] 鉴于上述存在的不足,针对现有的交通监测中存在的问题与缺陷,本发明提供了一种高分辨率卫星遥感交通流信息自动采集方法,方法易行,操作简便。 [0024] In view of the above-described deficiencies present, monitoring the existing traffic problems and defects, the present invention provides a high-resolution satellite remote sensing automatic traffic flow information acquisition method, the method easy, simple operation. 它不仅可用于大范围、快速、自动化采集道路网中车辆数量分布等静态交通流参数,还能够采集交通流速度、交通流量等动态参数。 It not only can be used for large-scale, rapid, automated collection of static road network traffic flow parameters such as the number of vehicles distributed, dynamic parameters can also collect traffic speed, traffic and so on. 并且由于卫星数据覆盖范围大(如QuickBird高分辨率卫星影像的幅宽达16.5km),而本发明充分考虑了大数据量处理的时间效率与自动化程度,提出的图像频率域相关匹配法与传统的空间域相关系数匹配法相比,时间效率提高了1.6倍。 And due to the large coverage satellite data (e.g., high-resolution satellite images QuickBird width of 16.5km), but the present invention fully contemplates the time efficiency and automation process large amounts of data, the proposed frequency domain correlation image matching with the conventional spatial domain correlation matching method compared to the time efficiency is improved 1.6 times.

[0025] 为了实现上述目的,本发明采用以下技术措施: [0025] To achieve the above object, the present invention adopts the following technical measures:

[0026] 其构思是: [0026] it is contemplated:

[0027] 1.本发明利用结合神经网络与面向对象分类方法对高分辨率卫星遥感影像进行道路区域内的车辆目标提取。 [0027] 1. The present invention utilizes neural network and remote sensing images of high-resolution satellite vehicle within the road area extracting a target object oriented classification.

[0028] 神经网络是一种模范动物神经网络行为特征,进行分布式并行信息处理的算法数学模型。 [0028] The neural network is one kind of network behavior characteristic neural model animal, a mathematical model of the algorithm information distributed parallel processing. 这种网络依靠系统的复杂程`度,通过调整内部大量节点之间相互连接的关系,从而达到处理信息的目的。 Such networks rely on the complexity of the process system ', by adjusting the relationship between the internal large number of interconnected nodes, so as to achieve the purpose of processing information.

[0029] 与传统遥感分类方法相比,神经网络分类方法具有自学习、自组织的能力,能最大限度地利用高分辨率卫星遥感影像中车辆样本集的先验知识,在无模型假设的前提下自动提取识别规则,从而提高了算法自动化处理能力。 [0029] Compared with the traditional remote sensing classification methods, neural network classification method has self-learning, self-organization capabilities, to maximize the use of high resolution remote sensing satellite image sample set of a priori knowledge of the vehicle, without model assumptions premise automatic extraction of identification rules, thereby improving the algorithm automation capabilities. 但其对光谱相似的车辆与阴影、路面标记之间的区分尚未达到最佳的效果。 But its vehicle with similar spectral shadow, the distinction between the pavement marking yet achieve the best results.

[0030] 面向对象遥感图像分类方法是以影像分割后若干像素组成的对象为基本操作单元,而非以灰度像元为单位。 [0030] Object-Oriented Image Classification is based on the segmented object image consisting of a number of pixels for the basic operation of the unit, rather than in gray-scale pixel units. 由于对象比单个像元更有实际意义,因此更能体现出高分辨率遥感影像与物理世界的联系,可以借此挖掘出更多的隐含信息;可以依据车辆对象的形状、面积等属性,将光谱相近的车辆目标与阴影、路面标记等区分开来。 Since the object pixel is more meaningful than a single, and therefore better reflect the high-resolution remote sensing image associated with the physical world, can take this dig out more implicit information; can be based on the shape and area of ​​the vehicle object's properties, the vehicle spectrum similar goals and shadow, pavement markings to distinguish.

[0031] 2.本发明利用图像频域率相关理论计算两幅影像的相关性。 Calculating a correlation of two images [0031] 2. The present invention utilizes the theory of frequency domain image. 计算公式如下: Calculated as follows:

[0032] [0032]

f\x,y) ο /?(.v, v) OF (w,v)H(ur) f \ x, y) ο /?(.v, v) OF (w, v) H (ur)

[0033] 其中,f与h为需要计算相关性的两幅影像,x、y分别为影像的行、列。 [0033] where, f and h is the need to calculate the correlation of two images, x, y are image row, column. F (u,v)和H (u,v)分别表示f (1,7)和11 (X,y)的傅里叶变换(U,V分别为影像行、列);F*是F的复共轭。 F (u, v) and H (u, v) represent f (1,7) and 11 (X, y) is the Fourier transform (U, V respectively the image row, column); F * is F, complex conjugate.

[0034] 根据图像处理的相关理论,空间域的相关性可由F* Cu, V) H Cu, V)的傅里叶反变换得到,而频域率中的计算可以大大减少计算的时间消耗,对同一遥感平台获取的具有拍摄时间差的全色和多光谱影像实现高效率的车辆匹配。 [0034] Based on the theory of image processing, spatial domain correlation by F * Cu, V) H Cu, V) of the inverse Fourier transform, and the rate is calculated in the frequency domain can greatly reduce the computing time consumption, match on the same vehicle platform acquired remote sensing has taken time difference panchromatic and multispectral images to achieve high efficiency. 本发明通过试验表明:采用不同的匹配方法对同一遥感图像进行匹配运算时,频率域相关匹配与空间域相关系数匹配法相t匕,时间效率提高了1.6倍。 The present invention results show: when the same image sensing operation using different matching matching method, the frequency domain correlation coefficient matches the spatial domain correlation matching dagger wears t, time efficiency is improved by 1.6 times.

[0035] 一种高分辨率卫星遥感交通流信息自动采集方法,包括以下步骤: [0035] A High Resolution Satellite automatic traffic flow information acquisition method, comprising the steps of:

[0036] 步骤一,包含道路的高分辨率卫星影像预处理,包括全色影像与矢量路网及全色与多光谱影像配准,道路区域分割和双边滤波增强。 [0036] Step a high-resolution satellite image preprocessing including road including the road network and vector panchromatic and multi-spectral and panchromatic image registration, the road region segmentation and enhance bilateral filtering. 配准包括矢量路网与全色遥感影像(即矢量与影像)以及同一卫星遥感平台同时获取的全色影像与多光谱影像(即影像与影像)的配准两个步骤。 The registration vector comprising the road network and panchromatic remote sensing image (i.e., image vector) and panchromatic and multi-spectral image of the same remote sensing platform acquired simultaneously (i.e., image and video) two registration steps. 全色影像与多光谱影像的配准精度一般应高于0.5个像素,矢量路网与遥感影像的配准精度不低于二个像素。 Panchromatic image and the registration accuracy multispectral images should generally be higher than 0.5 pixel, the accuracy of registration vectors road network for remote sensing is not less than two pixels. 道路区域分割即利用现有道路网的矢量数据(即矢量路网数据),以1/2的道路宽度为缓冲区半径,生成道路缓冲区;再以该缓冲区为模板对影像进行分割。 Road area dividing i.e. using vector data (i.e., vector network data) existing road network, road width to 1/2 of the radius of the buffer, the buffer generating a road; then the template image buffer is divided. 另外,对多光谱近红外波段选择合适的阈值进行影像分割,以去除带状植被区域,形成最终的道路影像以便下步进行车辆提取。 Further, the selection of an appropriate multispectral near infrared image segmentation threshold value, the vegetation region to remove the strip, so that the final image of the road at a vehicle extraction step. 此外,本发明采用了双边滤波方法对影像进行增强,在保证车辆边缘信息不丢失的情况下,可以尽可能地对道路面进行平滑,从而提高车辆分类的效率和效果。 Further, the present invention uses bilateral filtering method for image enhancement, edge in the case of the vehicle to ensure that information is not lost, the road surface can be smoothed as much as possible, thereby improving the efficiency and effectiveness of classification of the vehicle.

[0037] 步骤二,利用经过步骤一处理生成的道路区域遥感影像建立车辆遥感影像特征库。 [0037] Step two, through the use of a road area remote sensing image processing step of generating image features to establish vehicle telemetry library. 首先选取该影像中的小汽车、面包车、公交车等几种类型的典型车辆进行目视判读,计算其面积、长宽比、方向、矩形相似度等特征值,以此为样本,将这些特征值存入到数据库中,为后续车辆提取中目标阈值设定的依据。 First, the image is typically selected cars, vans, buses and other types of vehicle visual interpretation, calculated eigenvalues ​​area, aspect ratio, orientation, and other rectangular similarity, as a sample, these features values ​​are stored in the database, the extraction threshold value is set based on the target vehicle for follow-up.

[0038] 步骤三,对经过步骤一处理生成的道路区域全色影像进行结合神经网络与面向对象方法的车辆目标提取。 [0038] Step III road panchromatic image area generated through the processing performed in step a neural network and the target vehicle extracting object-oriented approach. 首先,选择典型的车辆样本(正样本)、非车辆样本(负样本)进行径向基神经网络训练,并利用训练好的网络进行车辆目标分类;随后对分类结果利用形态学闭运算进行车体的连接,以保证车辆目标对象特征的完整性,实现对候选车辆目标的粗略提取(称为“车辆粗提取”)。 First, a typical vehicle selected samples (positive samples), non-vehicle samples (negative samples) for RBF neural network training, the trained network and the vehicle target classification; subsequent classification results using morphological closing body connection, to ensure the integrity of the target object feature vehicle, the vehicle to achieve a coarse candidate extraction target (referred to as "vehicle crude extract"). 其次,将车辆粗提取结果影像分割为候选车辆对象和背景对象,并提取各对象特征(包括面积、长宽比、方向、矩形相似度等);接下来以车辆遥感影像特征库中的特征值作为车辆目标判定阈值,对上述候选车辆对象进一步进行自动筛选,去除相关干扰的地物(其中的方法包括利用宽度特征剔除道路标准线、利用方向特征剔除横跨道路的标志牌、利用面积特征剔除噪声点、利用矩形相似度排除其他地物等),从而得到精确的车辆目标(称为“车辆精提取”);最后输出车辆的总数量,单个车辆的重心坐标、面积、长度等属性信息。 Next, the crude extract vehicle as a result of image segmentation candidate vehicle and background objects, features and extracts each object (including area, aspect ratio, orientation, rectangular, similarity, etc.); next to the vehicle remote sensing image feature library eigenvalues determining a vehicle target threshold, the candidate vehicle the above objects further automatically filtered to remove the associated interference feature (excluding a road which process comprises standard line by using the feature width direction culling feature across a road sign, the use of culling feature area noise points, with a rectangular exclude other feature similarity, etc.) to obtain an accurate vehicle destination (referred to as "fine vehicle extract"); the total amount of the final output of the vehicle, the coordinates of the center of gravity of individual vehicles, area, and length attribute information.

[0039] 步骤四,在完成对步骤三中全色影像车辆检测及车辆目标提取的基础上,利用图像频率域相关匹配法,在多光谱影像中搜索对应车辆的具体位置。 [0039] Step 4 Upon completion of the step 3 and panchromatic vehicle detection object extraction on the vehicle, using a frequency domain correlation image matching method, the search for multispectral image corresponds to a specific location of the vehicle. 为了提高搜索效率。 In order to improve search efficiency. 本发明设定在较小的搜索窗口内进行图像匹配。 The present invention is set within a smaller image matching the search window. 根据车辆运动的特点,该搜索窗口要比全色影像中的车辆位置在车辆行驶方向上扩大一定的搜索范围,便于同一车辆的精确匹配。 According to the characteristics of the vehicle motion, the search window position of the vehicle in the full-color image enlarged certain search range in the vehicle traveling direction for precise matching than the same vehicle.

[0040] 步骤五,根据步骤三和步骤四得到的同一车辆的位置坐标,计算同一遥感平台中的全色与多光谱影像中同一车辆的位移量,再与这两种影像的拍摄时间差相除,即可计算出该车辆的行驶速度;根据车辆数量与交通流密度等参数的相关关系,根据步骤三得到的车辆数量与大小信息,计算道路区域的交通流密度,同时可统计出车辆类型;根据单个车辆的速度与交通流量、路段交通流速度等参数的关系,在以上参数获取的基础上,进一步计算出交通流量、交通流速度、道路占有率、车头距离等其他交通流参数。 [0040] Step 5 According to the same position coordinates of the vehicle obtained in step three and step four, the calculation of the amount of displacement of the same sensing platform PAN and multi-spectral image of the same vehicle, and then divided by the time difference between the two captured images can be calculated the speed of the vehicle; based on the correlation parameter and the number of vehicle traffic density, obtained in three steps according to the vehicle number and size information, calculating the density of the road traffic area, while the statistics of the type of vehicle; according to the parameters of the relationship between the speed of individual vehicles and traffic, road traffic flow speed, based on the above parameters obtained on further calculate other parameters of traffic flow traffic flow, traffic flow speed, share the road, the front distance.

[0041] 步骤六,利用人工识别的数据或者地面实测数据验证以上步骤得到的交通流参数,对该方法提取交通流信息进行精度评估。 [0041] Step six, or artificial ground Found identified data traffic flow parameters step verification data obtained above, the method of extracting traffic flow information for accuracy evaluation. 当所测路段被地面交通监测设备所覆盖时,可选择地面实测数据对同时期同路段的交通流参数进行对比验证;否则,可利用人工目视判读的方法,识别卫星影像中的车辆个数与车辆位移,计算人工识别的交通流参数,与本方法的结果进行对比验证。 When the measured surface is covered road traffic monitoring device, select ground truth data traffic flow parameters of the same period were comparatively sections; otherwise, the method may utilize artificial visual interpretation, vehicle identification number of satellite images calculating the traffic flow parameters manual identification, comparison with the results of verification of the method the displacement of the vehicle.

[0042] 在上述自动采集交通流信息的方法中,所采用的卫星遥感影像是同一遥感平台在一定时间差(大于0.2秒)获取的全色与多光谱影像。 [0042] In the method of automatic acquisition of traffic flow information, the satellite remote sensing image used is a full color image with the same multi-spectral remote sensing platforms in a predetermined time difference (greater than 0.2 seconds) obtained. 全色波段影像地面分辨率优于0.61m,多光谱波段影像的地面分辨率优于2.44m。 Ground resolution panchromatic image than 0.61m, multispectral bands ground resolution images than 2.44m.

[0043] 本发明与现有技术相比,具有以下优点。 [0043] Compared with the prior art the present invention has the following advantages.

[0044] 1、与传统的感应线圈、视频检测器采集交通流信息相比,本发明提出的利用卫星影像进行交通流信息采集流程方法可以应用于范围更大的道路网。 [0044] 1, the conventional induction coil, video capture detector compared to traffic flow information, provided by the invention using a satellite image process traffic information collection method can be applied to a larger range of the road network.

[0045] 2、与现有的遥感车辆检测技术方法相比,本发明实现了车辆目标分类器的阈值自动化选择,从而避免了人工干预阈值选择的主观性和阈值反复调整的复杂性,自动化程度更高,且能够自动、快速采集系列静态与动态的交通流信息。 [0045] 2, compared with the conventional remote sensing vehicle detection method of the present invention to achieve the threshold automated selection vehicle Classifier, thereby avoiding the complexity, automated manual intervention threshold selection subjectivity and the threshold value is repeatedly adjusted higher, and can automatically and quickly collect series of static and dynamic traffic flow information.

[0046] 3、与传统的空间域相关系数匹配法相比,本发明采用的图像频率域相关匹配法在车辆匹配的时间效率上提高了1.6倍。 [0046] 3, compared with traditional spatial domain correlation coefficient matching, a frequency domain correlation image matching method employed in the present invention, the time efficiency of the vehicle matches increased 1.6 times.

[0047] 4、本发明基于高分辨率卫星影像建立的自动化、高效率的交通流信息采集流程方法,可以实现车辆检测与计数、连续影像上(本说明中指一定拍摄时间间隔的全色与多光谱影像)的同一车辆目标的快速匹配,并进而获取交通流量、交通流速度、交通流密度、道路占有率、车头距离等静态和动态系列交通流信息,为交通信息自动采集提供了新的技术方法。 [0047] 4, the present invention is to establish a high-resolution satellite images based automation, efficient traffic flow information collection method, detection and counting of the vehicle can be achieved, the continuous images (full-color middle present description certain time interval and a multi-shot Quick match spectral images) of the same vehicle target, and thus get the series of static and dynamic traffic flow information traffic flow, traffic speed, traffic density, road share, from the front, to provide a new technology for traffic information collected automatically method.

[0048] 本发明在某城市区域5km*5km范围的美国QuickBird (快鸟)卫星影像上进行了交通流信息的提取实验,证明了该发明的有效性。 [0048] The present inventors have conducted experiments to extract traffic flow information in a city area 5km * 5km range of US QuickBird (QuickBird) satellite images to prove the effectiveness of the invention.

附图说明 BRIEF DESCRIPTION

[0049] 图1为一种高分辨率卫星遥感交通流信息自动采集方法流程示意图。 [0049] FIG. 1 as a high-resolution satellite remote sensing traffic information collected automatically flowchart of a method.

[0050] 图2为一种神经网络与面向对象相结合的遥感影像车辆检测流程示意图。 [0050] FIG. 2 is a neural network and the combination of object-oriented image sensing vehicle detection process FIG.

[0051] 图3为一种车辆匹配的流程示意图。 [0051] FIG. 3 is a flow diagram illustrating a vehicle match.

[0052] 图4为实验区域某路段的影像预处理结果示意图。 [0052] FIG. 4 is a schematic diagram of an image pre-processing results of the experiment a section of the area.

[0053] 其中4 Ca)为全色影像,4 (b)为多光谱影像。 [0053] wherein 4 Ca) full-color image, 4 (b) is a multi-spectral image.

[0054] 图5为实验区域某路段的车辆粗提取结果示意图。 [0054] FIG. 5 is a section of a schematic of the experimental area of ​​the vehicle crude extraction result.

[0055] 图6为实验区域某路段的车辆精提取示意图。 [0055] FIG. 6 is a schematic view of the extraction experiments for a road vehicle fine region.

[0056] 其中,图6(a)为车辆粗提取结果中待剔除目标,6(b)为车辆精提取结果,6 (C)为叠加全色影像后的车辆目标范围。 [0056] wherein, FIG. 6 (a) results in a crude extract of the vehicle target to be deleted, 6 (b) is a fine vehicle extraction result, 6 (C) is the target range of the superimposed vehicle panchromatic image.

[0057] 图7为实验区域中某路段局部的车辆匹配结果示意图。 [0057] FIG. 7 is a partial match for a road vehicle showing the results of the experimental area.

[0058] 7 (a)中方框为全色影像中车辆,图7 (b)方框为通过匹配搜索到的多光谱影像中的对应车辆。 [0058] 7 (a) is a block in the full-color image of the vehicle, FIG. 7 (b) is a block corresponding to the vehicle searched by matching the multispectral image.

[0059] 图8为实验区域全部路段的车辆检测结果示意图。 [0059] FIG. 8 is a schematic diagram of the vehicle detection result of the experiment all area sections.

[0060] 图9为采用本发明方案采集的实验区域各交通流参数信息示意图。 [0060] FIG. 9 is a regional program of the present invention using experimental acquisition parameter information for each traffic flow schematic.

[0061] 其中,图9 (a),9 (b),9 (c),9 (d),9 (e)和9 (f)分别为路段车辆数目、交通流密度、车头距离、道路占有率、平均速度以及交通流量。 [0061] wherein FIG. 9 (a), 9 (b), 9 (c), 9 (d), 9 (e) and 9 (f), respectively, from the number of road vehicles, traffic density, the front road occupies rate, average speed and traffic flow. 具体实施方式 Detailed ways

[0062] 实施例1: [0062] Example 1:

[0063] 下面结合附图和具体实施案例对本发明做进一步详细描述(如图1所示)。 [0063] The present invention will be further described in detail (Figure 1) in conjunction with the accompanying drawings and the detailed description cases.

[0064] 一种高分辨率卫星遥感交通流信息自动采集方法,包含以下步骤。 [0064] A High Resolution Satellite automatic traffic flow information acquisition method, comprising the following steps.

[0065] 1.预处理101: [0065] 1. Pretreatment 101:

[0066] 为了提取卫星遥感影像中的有用信息,需要对输入的高分辨率卫星影像I进行预处理工作。 [0066] In order to extract useful information satellite remote sensing image, the need for preprocessing of the input image I high-resolution satellite. 本发明所包含的预处理包括:全色影像与矢量路网以及全色与多光谱影像的配准1011、道路区域分割1012、双边滤波影像增强1013。 The present invention comprises a preprocessing comprises: registration and Vector 1011 Panchromatic road network and the panchromatic and multi-spectral image, the area dividing road 1012, 1013 bilateral filtering image enhancement. 这三项工作处理对象为同一高分辨率遥感卫星平台在一定时间差(大于0.2秒)获取的高分辨率全色影像和多光谱影像(如QuickbirdX These three work processed high resolution panchromatic image and the same multi-spectral resolution remote sensing satellite platform at a certain time difference (greater than 0.2 seconds) acquired (e.g. QuickbirdX

[0067] 利用相同区域中同一遥感平台获取的具有拍摄时间差的高分辨率卫星影像1,进行全色影像与矢量路网及全色与多光谱影像的配准1011,即先在路网矢量数据的支持下,对全色影像进行位置配准,再以其中的全色影像为参考,对多光谱影像进行配准,以使两幅影像中的同名点具有相同的地理坐标。 [0067] In the same area using the same high-resolution satellite remote sensing platform obtains images having a difference in photographing time, registering a full-color image and vector 1011 and the road network PAN and multispectral images, i.e., the first vector data in the network with the support of panchromatic image registration position, and then the full color image which is the reference, the multi-spectral image registration, so that the points of the two images of the same name have the same geographical coordinates. 前者是为道路区域分割做准备;后者是为了保证车辆在图像上的偏移是由其自身的运动而不是影像的错位而引起的,以此来提高车辆位移计算的准确性。 The former is to prepare a road area division; the latter is to ensure that the vehicle on the image shift is offset by the movement itself, rather than the image caused, in order to improve the accuracy of displacement calculation of the vehicle. 其方法是分别在两种数据上选择同一地物的特征点(如道路中心线的拐点、端点、交叉点等),然后利用这些特征点建立两种数据之间的几何畸变模型(如多项式模型);利用该几何畸变模型进行几何校正,达到精确配准的目的。 The method is to select feature points, respectively (e.g., an inflection point of the road centerline, endpoints, intersections, etc.) of the same feature in the two types of data, are then used to establish the geometric distortion model feature points between the two data (e.g., polynomial models object model by using the geometric correction of geometric distortion, to achieve accurate registration;).

[0068] 道路区域分割1012是利用已知矢量路网数据2,以1/2的道路宽度为缓冲半径,生成道路缓冲区,再以缓冲区为模板对道路影像进行分割。 [0068] The area dividing path 1012 using a known vector network data 2, the road width to a radius of 1/2 of the buffer, the buffer generating a road, then the buffer is a template image of a road is divided. 此外,由于植被在近红外波段高反射,因此对多光谱影像的近红外波段选择合适的阈值进行影像分割,去除道路中的绿化带,形成最终的道路区域影像。 Further, since the vegetation in the near infrared high reflectance, and therefore the selection of near-infrared bands multispectral images suitable image segmentation threshold value, removing the green belt in the road, the road to form the final image region.

[0069] 由于路面油污造成路面光谱不均匀,影响车辆与路面的分割,因此本发明采用双边滤波影像增强1013方法对影像进行增强处理,在保证车辆边缘信息不丢失的情况下,尽可能平滑道路面,提高车辆分类的效率和效果。 [0069] Since the uneven road surface causes oil road spectrum, influence of the division of the vehicle and the road surface, thus the present invention uses image intensifier 1013 bilateral filtering method for image enhancement processing, in a case where the edge of the vehicle to ensure that information is not lost, as smooth as possible road surface, to improve the efficiency and effectiveness of vehicle classification.

[0070] 该步骤的结果如图4所示,其中,4a为预处理后的全色影像,4b为预处理后的多光谱影像。 [0070] The result of this step is shown in Figure 4, wherein, 4a is pretreated panchromatic, 4b multispectral images after pretreatment.

[0071] 2.车辆遥感影像特征库102: [0071] The vehicle remote sensing image feature database 102:

[0072] 首先利用经过图像预处理生成的道路区域全色遥感影像,对不同类型的典型车辆进行目视判读,再计算其面积、长宽比、方向、矩形相似度等特征值,以此为样本,将这些特征值存入到数据库中,为后续车辆提取中目标阈值设定的依据。 [0072] Firstly, a road area panchromatic image sensing after pre-generated images, the different types of vehicle typical visual interpretation, then the area is calculated characteristic values, aspect ratio, orientation, and other rectangular similarity, as a sample, these feature values ​​stored in the database, based on the target threshold value is set for a subsequent extraction of the vehicle. 车辆遥感影像特征库102是由微型车(长度〈3.5m)、轻型车(长度3.5〜7m)、中型车(长度7_10m,如)、大型车(>10m,如公交车、卡车)等不同类型车辆的长、宽、面积、长宽比、方向、矩形相似度等特征值及其属性名称与代码构成。 Vehicle remote sensing image feature database 102 by mini car (length <3.5m), light vehicle (length 3.5~7m), the car (length 7_10m, such as), large vehicles (> 10m, such as buses, trucks) of different types length, width, area, aspect ratio, orientation, rectangular similarity feature value and the like of the vehicle and codes constituting the attribute name.

[0073] 3.车辆提取103: [0073] The vehicle 103 Extraction:

[0074] 车辆提取包括神经网络车辆粗提取1031与面向对象车辆精提取1032两个阶段。 [0074] The vehicle includes a neural network to extract a crude extract of the vehicle 1031 extracts the object-oriented fine vehicle 1032 in two stages. 在面向对象车辆精提取中,需要根据事先通过目视判读高分辨率遥感影像建立的车辆遥感影像特征库102,利用其中的特征值作为阈值,参与面向对象车辆精提取。 In object-oriented fine vehicle extraction, by the need to advance the visual interpretation of the vehicle resolution remote sensing image created remote sensing image feature database 102, using the feature value as a threshold value which involved extraction of object-oriented fine vehicle.

[0075] 具体步骤如图2所示:[0076] ( I)神经网络车辆粗提取A: [0075] DETAILED Step 2: [0076] (I) The crude vehicle neural network Extract A:

[0077] 对步骤预处理101处理后的道路区域全色影像,选择一定数量(3-10个或10个以上)的正负样本Al进行神经网络训练A2,其中,正样本指车辆样本,负样本指非车辆样本。 [0077] The pretreatment step road panchromatic image region after processing 101, select certain number (3-10, or 10 or more) of the positive and negative samples for neural network training Al A2, wherein a sample of a vehicle positive samples, negative sample refers to non-vehicle samples. 由于车辆目标光谱特征明显,因此样本数量不需要选择太多,本实验各选取三个正负样本,即可训练得到分类效果较好的网络。 Since the vehicle target spectral characteristics significantly, and therefore the number of samples does not need too many choices, the selected three positive and negative samples of each experiment, to train a network to obtain better classification results. 再利用训练好的网络进行神经网络分类A3,分类得到车辆目标。 After using these networks neural network classification A3, obtained by classifying vehicle target. 本发明采用收敛性好、自学习能力强的径向基(RBF)神经网络进行车辆目标分类。 The present invention employs a good convergence, a vehicle from the target classification learning ability strong radial basis (RBF) neural network. 但该过程有可能造成不匀质的车体被分为几个部分,如拖车的车头与车厢相分离,光谱特征不同的车窗与车体相分离。 But the process may result in uneven quality of the vehicle body is divided into several parts, such as front and trailer separated compartments, different spectral characteristics window and the vehicle body separated. 对此,分类后利用形态学运算A4进行车辆的合并,以保证对象特征的完整性。 In this regard, the classification using morphological operations to merge A4 vehicle, to ensure the integrity characteristics of the object.

[0078] 对实验区域的其中一个道路区域进行神经网络车辆粗提取的结果,如图5所示,其中白色斑块为提取出的车体,黑色为背景。 [0078] Experimental region wherein a region of the road vehicle the neural network results in a crude extract, as shown in Figure 5, wherein the extracted white plaque vehicle body, a black background.

[0079] (2)面向对象车辆精提取B: [0079] (2) object-oriented vehicle sibiricum B:

[0080] 对神经网络车辆粗提取A结果进行影像分割BI,即将分类结果影像分割为候选车辆对象和背景对象,并进行对象特征提取B2。 [0080] A crude extract results of the image segmentation BI neural network of the vehicle, i.e. the classification result image into a candidate vehicle and background objects, and object feature extraction B2. 需要提取的特征包括对象面积、长宽比、方向、矩形相似度等。 Need to extract features includes an object area, aspect ratio, orientation, and the like rectangular similarity. 根据车辆遥感影像特征库102中的车辆特征(如车辆对象面积、长宽比、方向、矩形相似度等),分别利用这些特征值作为综合判定候选车辆目标的阈值,对候选车辆对象进一步筛选,去除干扰的地物,包括利用宽度特征剔除道路标准线、利用方向特征剔除横跨道路的标志牌、利用面积特征剔除噪声点、利用矩形相似度排除其他地物等,从而完成车辆对象识别B3 ;最后输出车辆对象参数4,包括车辆的总数量,单个车辆的重心坐标、面积、长度等属性信息。 The vehicle characteristics (e.g., the vehicle target area, aspect ratio, orientation, similarity rectangular, etc.) of the vehicle remote sensing image feature database 102, respectively, using the determined characteristic values ​​as an overall vehicle target candidate threshold, the vehicle further screening candidate objects, interference removal feature, including the use of standard line width of the road feature culling, characterized removed using the direction across the road sign, using the area excluding noise characteristic points, with a rectangular similarity exclude other surface features, etc., thereby completing the object recognition B3 vehicle; Finally, the object parameters output 4 of the vehicle, the vehicle including a total number of individual coordinates of the center of gravity of the vehicle, area, and length attribute information.

[0081] 对实验区域中某个道路区域进行面向对象车辆精提取的过程如图6所示,利用不同的对象特征,去除粗分类结果中的非车辆对象(如图6a中各框内的白色斑块)。 [0081] The experimental area region oriented to a road vehicle finishing process objects extracted as shown in FIG 6, characterized in using different target, removing non-vehicle objects rough classification results (each white box in Figure 6a plaque).

[0082] 采用本发明提出的车辆提取方法,比单纯采用采用面向对象的方法(谭衢霖等2008年专利中所提的方法),大大减少了对人工参数选择的依赖性。 [0082] The present invention proposed a method to extract vehicle, than a simple method (Method TAN Qu et Lin patent mentioned in 2008) object-oriented, greatly reducing the dependence on the parameters of artificial selection. 具体表现为两点:一是以径向基(RBF)神经网络的分类结果(二值分割图像)作为面向对象分类的数据输入,不需要再人工反复选择合适的尺度分割影像;二是以车辆遥感影像特征库中的特征阈值作为参考,进行车辆精提取,避免了人工调整参数阈值的主观性与复杂性。 Specific performance by two points: First, radial basis (RBF) neural network classification result (binary divided image) as input data for the object classification, repeated manual do not need to select the appropriate scale image segmentation; Second vehicle characterized in remote sensing threshold as a reference signature database, a vehicle sibiricum, avoiding manual adjustment parameter threshold subjectivity and complexity.

[0083] 因此,本发明提出的车辆提取方法提高了大范围路网遥感影像中车辆提取的自动化程度和时间效率。 [0083] Thus, the vehicle of the present invention proposed extraction method increases the degree of automation and time-efficient large-scale road network remote sensing image extracted from the vehicle.

[0084] 4.车辆匹配104: [0084] 4. The vehicle match 104:

[0085] 该步骤目的为根据全色影像中提取的车辆目标,利用图像频率域相关匹配方法,通过在多光谱影像中设定合适的搜索窗口搜索对应车辆的具体位置。 [0085] The steps according to the vehicle target object extracted from the panchromatic image, correlation in the frequency domain with the image matching method, by setting an appropriate search window in a multi-spectral image search corresponding to the specific position in the vehicle. 由于多光谱影像和全色影像经过了精确配准,同一车辆在传感器短暂的拍摄延迟内,只移动了很小的距离(例如,车辆行驶速度为100km/h,则在0.2s的拍摄延迟内,只移动了5.6m,表现在0.61m分辨率的卫星影像上,为9个像素),因此,为提高搜索效率,本发明设定在较小的搜索窗口内进行图像匹配。 Since the multi-spectral image and a panchromatic image and, after precise registration of the same vehicle within the imaging sensor short delay, only moved a small distance (e.g., vehicle traveling speed is 100km / h, in the imaging retardation 0.2s , moved only 5.6m, 0.61m performance on satellite image resolution of nine pixels), and therefore, in order to improve search efficiency, the present invention is set within a smaller image matching the search window. 根据车辆运动的特点,该搜索窗口要比全色影像中的车辆位置在车辆行驶方向上扩大一定的搜索范围,便于同一车辆的精确匹配。 According to the characteristics of the vehicle motion, the search window position of the vehicle in the full-color image enlarged certain search range in the vehicle traveling direction for precise matching than the same vehicle. 通过试验,一般应使该窗口在车辆行驶方向上扩大六m (地面距离)。 By experiment, the window should be expanded generally six m (from the ground) in the vehicle traveling direction.

[0086] 车辆匹配104的具体步骤如图3所示。 [0086] DETAILED vehicle match step 104 is shown in Fig. [0087] (I)将车辆提取103中的车辆区域作为模板影像f(x,y)6,以多光谱影像中对应的搜索窗口作为目标影像h(x,y)9分别进行傅里叶变换,转化到频率域,得到对应的频率域图像F(x,y) 7和频率域图像H (X,y) 10。 [0087] (I) in the vehicle to extract the vehicle 103 as a template region image f (x, y) 6, a multi-spectral image as a target search window corresponding to an image h (x, y) 9, respectively Fourier transform , transformed to the frequency domain, to obtain the corresponding frequency domain image F (x, y) 7 and the frequency domain image H (X, y) 10.

[0088] (2)将频率域图像F(x,y)7的复共轭图像F*(x, y) 8与频率域图像H(x, y) 10相乘,得到乘积图像11。 [0088] (2) The complex frequency domain image F (x, y) 7 conjugate image F * (x, y) 8 multiplied by the frequency domain image H (x, y) 10, the image 11 to obtain the product.

[0089] (3)对乘积图像11进行傅里叶逆变换,得到模板影像与目标影像的相关图像g(x, y) 12,相关图像中的最大值所在的行列值,即模板影像与目标影像达到最佳匹配时的位置。 [0089] (3) the product of an inverse Fourier transform image 11 to obtain the target image and the template image related image g (x, y) 12, the ranks of the maximum of the correlation value image is located, i.e. the target image and the template when the image reaches the position of the best match.

[0090] 利用上述方法对实验区域的中一个路段行车辆匹配,结果如图7所示。 [0090] With the above-described method in a road vehicle test line area match, the results shown in Fig. 其中,图7a中方框为全色影像中的车辆,图7b方框为通过车辆匹配搜索到的多光谱影像中的对应车辆。 Wherein FIG. 7a is a panchromatic image block in a vehicle, FIG. 7b is a block corresponding to the vehicle through a vehicle matching the searched multi-spectral image. 对于实验区域的车辆匹配,车辆定位误差平均约为2m,标准方差约为2m。 Experimental area match the vehicle, vehicle positioning error averaging about 2m, the standard deviation is about 2m. 其中约有18%的匹配点误差超过5m,约68%的匹配点误差在三m以内。 Wherein about 18% of the matching point error exceeds 5m, about 68% of errors within three matching points m. 该精度可以满足大范围宏观交通调查的需要。 The precision meet the needs of a wide range of macro-traffic investigation.

[0091] 5.交通流参数估算105: [0091] The traffic flow parameter estimator 105:

[0092] 在车辆提取103和车辆匹配104的基础上,估算单个车辆行驶速度、路段交通流速度、交通流密度、道路空间占有率、车头距离等交通流参数,从而实现交通流信息的自动采集。 [0092] In the vehicle 103 and the vehicle match the extraction on the basis of 104, estimated vehicle speed single, traffic flow parameters Traffic Flow velocity, traffic density and road space occupancy, from the front, so as to achieve automatic acquisition of traffic flow information . 各参数的具体计算方法如下。 Specific calculation method parameters are as follows.

[0093] ( I)单个车辆行驶速度计算: [0093] (I) a single vehicle running speed is calculated:

[0094] 道路上行驶的单个车辆的行驶速度(V)可由下列公式进行计算,单位m/s (或km/h)。 Individual vehicles traveling on a [0094] road speed (V) is calculated by the following formula, units of m / s (or km / h).

[0095] [0095]

Figure CN102855759BD00121

[0096] 其中,(xp,yp)与(xm,yffl)分别为全色与多光谱影像上对应车辆重心的像素行列值,由步骤车辆提取103和步骤车辆匹配104分别得出;r为影像地面分辨率(全色与多光谱影像均采样为同一分辨率),单位为m (或km);t为全色与多光谱拍摄时间间隔,单位为s (或h)。 [0096] wherein, (xp, yp) and (xm, yffl) are full-color and multi-spectral image corresponding to the ranks of the pixel values ​​of the center of gravity of the vehicle, the vehicle is extracted by the step 103 and step 104 matches the vehicle are obtained; r is an image ground resolution (panchromatic and multi-spectral image are the same sampling resolution), in units of m (or km); t is the panchromatic and multi-spectral imaging time interval, s (or h).

[0097] (2)路段交通流速度计算: [0097] (2) Calculation of Traffic Flow velocity:

[0098] 在连续的自由车流条件下,路段交通流速度可用区间平均速度U来表示,单位m/s (或km/h): [0098] In the continuous free traffic conditions, road traffic flow velocity U available interval average speed expressed in units m / s (or km / h):

Figure CN102855759BD00122

[0100] 其中M为路段车辆数量(veh),由步骤车辆提取1032得到;Vi指第i车观测时的瞬时速度(m/s或km/h)。 [0100] where M is the number of road vehicle (veh), extracted by the vehicle obtained in step 1032; Vi means the instantaneous speed (m / s or km / h) car when the i-th observation.

[0101] (3)交通流密度计算: [0101] (3) traffic density calculation:

[0102] 交通流密度(k)表示交通流的疏密程度,即一定单位长度的道路上所有车辆的数量,单位为(veh/km),可表示为: [0102] traffic density (k) represents the degree of traffic flow density, i.e. the number of all vehicles on the road a certain unit length, in units of (veh / km), can be expressed as:

[0103] [0103]

Figure CN102855759BD00123

[0104] S为测量道路段的长度(km),根据已有的路网矢量数据获得;M为道路段中的车辆数量(veh),由步骤车辆提取103得到。 [0104] S is the measured length (km) of road segments, according to the existing road network data obtained vectors; M is the number obtained in the road segment the vehicle (VEH), vehicle extracted by step 103.

[0105] (4)交通流量计算: [0105] (4) traffic Calculated:

[0106] 在已知交通流速度与交通密度的前提下,交通流量q可表示为: [0106] Under the premise of a known traffic speed and traffic density, traffic flow q can be expressed as:

[0107] q = kXu [0107] q = kXu

[0108] 即交通流量(q)等于交通密度(k)与区间平均速度(U)的乘积,单位为veh/h (或veh/s)。 [0108] i.e. the traffic flow (q) is equal to the product of the traffic density (k) and the average interval velocity (U), in units of veh / h (or veh / s).

[0109] (5)道路空间占有率计算: [0109] (5) road space occupancy calculation:

[0110] 道路空间占有率(Rs)表示单车道上车辆总长度和路段总长度之比(%): [0110] road space occupancy (Rs) represented by the ratio of the total length and total length of the single lane road vehicle (%):

[0111] [0111]

Figure CN102855759BD00131

[0112] 其中S为观测路段长度,根据已有的路网矢量数据获得;Si表示第i辆车的长度,由步骤车辆提取103获得。 [0112] where S is the observed length of the link, according to the existing road network data obtained vectors; Si represents the i-th length of the vehicle, the vehicle extracted at step 103 is obtained.

[0113] (6)车头距离计算: [0113] (6) the front distance calculation:

[0114] 车头距离包括空间与时间上的两种形式一车头间距与车头时距。 [0114] includes a spatial distance from the front when the two forms on the front and a headway time. 车头间距表示在同向行驶的一列车队中,相邻两辆车的车头之间的距离;车头时距表示在同向行驶的一列车队中,相邻两辆车驶过同一个断面的时间差。 Represents the same pitch in the front of a train traveling team, the distance between the front two adjacent vehicle; represents a distance in the traveling direction with a convoy, the neighboring two vehicle passes the same time when the difference between the front section. 路段中所有车头间距的平均值为平均车头间距(Hs),单位(m\eh),表示为: The average of all segment Headway average headway (Hs), the unit (m \ eh), expressed as:

Figure CN102855759BD00132

[0116] 式中:Ksingle为单车道交通流密度(veh/km)。 [0116] wherein: Ksingle lane traffic density is (veh / km). 同理,平均车头时距(Ht)与交通量的 Similarly, the average headway (Ht) and the amount of traffic

关系用如下公式表示,单位(s/veh): Relationship expressed by the following equation, the unit (s / veh):

Figure CN102855759BD00133

[0118]其中,Qsingle 为交通流量(veh/h )。 [0118] wherein, Qsingle of traffic flow (veh / h).

[0119] 图8中方框为实验区域内八个路段的车辆检测结果。 [0119] FIG. 8 is a block eight sections of the vehicle within the experimental area detection result. 其车辆检出率均达到90%以上。 Detection rate of the vehicle which more than 90%. 图9为采用本发明方案采集的实验区域内的各交通流参数信息。 9 is adopted for each traffic flow parameters in the information area experimental embodiment of the present invention acquired.

[0120] 6.精度评估106: [0120] 6. Accuracy Evaluation 106:

[0121] 由于目前从卫星遥感数据自动获得的交通流信息难免存在误差,因此,利用人工识别或地面监测获得的数据3对结果进行精度评估当所测路段被地感线圈、摄像头等地面交通监测设备所覆盖时,可利用这些设备获取的车辆计数、车辆行驶速度以及整路段的交通流参数数据,与同时期同路段的卫星影像获取的交通流参数进行对比验证;否则,可利用人工目视判读的方法,识别卫星影像中的车辆个数与车辆位移,计算人工识别的交通流参数,与本方法的结果进行对比验证。 [0121] Due to the current traffic flow information is obtained automatically from the satellite remote sensing data inevitable errors, and therefore, the data using the manual identification or ground monitoring 3 obtained results for accuracy evaluation when the measured link is sense coil, cameras ground traffic monitoring when the device is covered, the traffic flow parameters available data from these vehicle devices acquired count, vehicle speed, and the whole road section, traffic flow parameters acquired satellite image of the same period were comparatively sections; otherwise, may be visually artificial the method of interpretation, the displacement of the vehicle identification number of the vehicle in satellite images, calculates the traffic flow parameters manual identification, validation and comparing the results of the method.

[0122] 以上所述,为本发明的具体实施方案。 [0122] The above specific embodiments of the present invention. 根据本发明,可以从高分辨率(全色波段分辨率优于0.61m)卫星影像中,通过神经网络与面向对象相结合的方法自动提取全色影像道路中行驶的车辆目标;再利用图像频率域相关匹配法,实现全色与多光谱影像中相对应的车辆匹配定位,得到车辆数量、长度以及这两种影像中的车辆重心坐标等信息,最后根据此计算出道路区域的交通流速度、密度、交通流量、道路空间占有率、车头距离等交通参数。 According to the present invention, (0.61m better resolution panchromatic) Satellite images from the high resolution, full-color image to automatically extract a target vehicle traveling road by a neural network for combining with the object; the image frequency reuse domain correlation matching method, full-color and multi-spectral image corresponding to the positioning of the vehicle match, the number of vehicles to obtain information, as well as the length of the vehicle center of gravity coordinates in the two images, the last traffic flow based on this calculated road speed region, the density of traffic parameters, traffic flow, share road space, front distance. [0123] 本发明所用方法具有大范围、快速、自动化采集交通流信息的优势,且与现有遥感交通流信息采集方法相比,采集的交通流信息覆盖范围更广,数据处理的自动化程度和处理效率也得到提高。 [0123] The present invention has a wide range of the method, rapid, automated advantage traffic information collection, and is compared with the conventional method of sensing traffic information collection, traffic flow information acquired broader coverage, degree of automation and data processing processing efficiency is also improved. 本发明技术实用性强,可在宏观交通监测和管理中推广应用。 The present invention practical technology, can promote the use of macro traffic monitoring and management.

[0124] 以上实施方案可以使本领域技术人员更全面理解本发明,但本发明的保护范围并不局限于此,一切不脱离本发明的精神和技术实质的技术方案及其改进,均应涵盖在本发明专利的保护范围之内。 [0124] the above embodiments may enable those skilled in the art to more fully understand the invention, but the scope of the present invention is not limited thereto, all without departing from the spirit and essence of the technical disclosure of the present technical solutions and improvements should fall within the scope of the present invention patent.

Claims (1)

1.一种高分辨率卫星遥感交通流信息自动采集方法,其步骤是: A、预处理(101): 为了提取卫星遥感影像中的有用信息,对输入的高分辨率卫星影像(I)进行预处理,包含的影像预处理(101)工作包括:全色与多光谱影像以及影像与矢量路网的配准(1011)、道路区域分割(1012)、双边滤波影像增强(1013),这三项工作处理对象为同一传感器的高分辨率全色和多光谱影像; 全色与多光谱影像以及影像与矢量路网的配准是将已知的矢量路网数据(2)与高分辨率卫星影像(I)进行匹配,并将同一传感器获取的全色与多光谱影像进行匹配,前者是为了车辆在图像上的偏移是由其自身的影像的错位引起,来提高车辆位移计算的准确性;后者是为下步的道路区域分割做准备,其方法是分别在两种数据上选择同一地物的特征点:道路中心线的拐点、端点、交叉点,然后利用特征点建立 A high-resolution remote sensing automatic traffic flow information acquisition method, the steps are: A, pretreatment (101): In order to extract useful information in satellite remote sensing image, high-resolution satellite imagery (I) inputted pretreatment, pretreatment comprising the image (101) include: registering (1011) with a multi-spectral image and a panchromatic image and vector road network, road region segmentation (1012), image enhancement bilateral filtering (1013), three work item is processed the same panchromatic and multi-spectral image sensors; registration PAN and multispectral image and the image and the road network are vector known vector network data (2) and the high-resolution satellite image (I) by matching the panchromatic and multi-spectral image acquired by matching the same sensor, the former is offset to the vehicle on the image by its own image caused by misalignment, to improve the accuracy of the calculated displacement of the vehicle ; the latter is to prepare for the next step of dividing the road region which is selected feature points are the same feature in the two kinds of data: the inflection point of the road centerline, endpoints, intersections, and then using the feature point establishes 种数据之间的几何畸变模型;利用几何畸变模型进行几何校正,达到精配准; 道路区域分割(1012)是利用已知道路矢量路网数据,以1/2的道路宽度为缓冲半径,生成道路缓冲区,再以缓冲区为模板对影像进行分割,植被在近红外波段高反射,对多光谱影像的近红外波段选择阈值进行影像分割,去除道路中的绿化带,形成最终的道路影像以便下步进行车辆提取; B、车辆提取(103): 车辆提取包括神经网络车辆粗提取(1031)与面向对象车辆精提取(1032)两个阶段,在面向对象车辆精提取中,根据车辆遥感影像特征库(102),利用其中的特征值作为阈值,进行面向对象车辆精提取; (1)利用神经网络方法进行车辆粗提取(A): 对步骤预处理(101)处理后的道`路段全色影像,选择3-10个的正负样本(Al)进行神经网络训练(A2),其中,正样本指车辆样本,负样本指非车辆样本,再 Geometric distortion between the model data types; geometric model using the geometric distortion correction, to achieve fine registration; road area dividing (1012) the vector using a known road network data, the road width to a radius of 1/2 buffer to generate road buffer, then the video buffer as the template is divided, high vegetation near infrared reflection, the selection of near-infrared bands of the multispectral images image segmentation threshold value, removing the green belt in the road, the road image to form a final the next step for extraction of the vehicle; B, extracting the vehicle (103): extracting a vehicle comprising a vehicle crude extract neural network (1031) with extraction (1032) refined in two stages the object-oriented vehicle, refined object-oriented vehicle extraction, according to the vehicle remote sensing images characteristic feature database (102), using which the value as a threshold value, object oriented vehicle sibiricum; (1) a vehicle using neural network crude extract (a): a step of preprocessing (101) tract `link the processed whole color images by the positive and negative samples 3-10 (Al) for training the neural network (A2), wherein the sample positive samples of a vehicle, the vehicle refers to a non negative samples sample, and then 利用训练好的网络进行神经网络分类(A3),分类得到车辆目标,采用收敛性好、自学习能力强的径向基神经网络进行车辆目标分类,分类后利用形态学运算(A4)进行车辆的合并,以保证对象特征的完整性; (2)利用面向对象方法进行车辆精提取(B): 对利用神经网络方法进行车辆粗提取(A)结果进行影像分割(BI),将分类结果影像分割为候选车辆对象和背景对象,并进行对象特征提取(B2),提取的特征包括对象面积、长宽t匕、方向、矩形相似度;根据目视判读的车辆影像特征:车辆对象面积、长宽比、方向、矩形相似度,所建立的车辆遥感影像特征库(102),分别利用其中的特征值作为综合判定车辆的值域,对候选车辆对象进一步筛选,并去除干扰的地物,完成车辆对象识别(B3);最后输出车辆对象参数(4),包括车辆的总数量,单个车辆的重心坐标、面积、长度属性 Using the trained neural network classifiers network (A3), get vehicle target classification, the use of convergence, and strong ability to learn from radial basis function neural network classification target vehicle, a vehicle after classification using morphological operations (A4) of the combined to ensure the integrity of the object feature; (2) object-oriented method of extracting a vehicle finish (B): on a vehicle by using neural networks crude extract (a) the results of the image segmentation (the BI), the classification result image segmentation for the candidate vehicle and background objects, and object feature extraction (B2), the extracted feature includes an object area, t dagger length and width directions, a rectangular similarity; vehicle according to visual interpretation of the image characteristic: the vehicle target area, length and width ratio, direction, rectangular similarity vehicle established remote sensing image feature database (102), respectively, using the feature value as a range where the comprehensive determination of the vehicle, the vehicle further screening of candidate objects, and interference removal feature, complete vehicle object identification (B3); final output object parameters of the vehicle (4), comprising a total number of vehicles, the coordinates of the center of gravity of individual vehicles, area, length attribute 信息;C、车辆匹配(104): 该步骤目的为根据全色影像中提取的车辆目标,利用图像频率域相关匹配方法,通过在多光谱影像中设定搜索窗口搜索对应车辆的具体位置,多光谱影像和全色影像经过了精确配准,同一车辆在传感器短暂的拍摄延迟内,只移动了很小的距离,为提高搜索效率,设定在搜索窗口内进行图像匹配,根据车辆运动的特点,该搜索窗口要比全色影像中的车辆位置在车辆行驶方向上扩大搜索范围,便于同一车辆的精确匹配,通过试验,窗口在车辆行驶方向上扩大六米,具体步骤如下: (1)将车辆提取(103)中的车辆区域作为模板影像f(x,y) (6),以多光谱影像中对应的搜索窗口作为目标影像h(x,y) (9)分别进行傅里叶变换,转化到频率域,得到对应的频率域图像F(x,y) (7)和频率域图像H(x, y) (10); (2)将频率域图像F(x,y) (7)的复共轭图像F*(x,y) (8) Information; C, matching the vehicle (104): This step is the vehicle target object extracted from the panchromatic image, correlation in the frequency domain with the image matching method, the specific location of the vehicle by setting a search window in the search for the corresponding multi-spectral image, multiple spectral image and a panchromatic image and, after precise registration of the same vehicle within the imaging sensor short delay, only moved a small distance, to improve the efficiency of the search, setting the image match within the search window, according to the characteristics of vehicle motion , the search window position of the vehicle than the panchromatic image is enlarged in the vehicle traveling direction search range for precise matching of the same vehicle, by experiment, six meters expanded window in the vehicle traveling direction, the following steps: (1) vehicle region extraction vehicle (103) as a template image f (x, y) (6), (9) are Fourier transformed to the corresponding multi-spectral image as a target search window image h (x, y), transformed to the frequency domain, to obtain the corresponding frequency domain image F (x, y) (7) and the frequency domain image H (x, y) (10); (2) the frequency domain image F (x, y) (7) complex conjugate image F * (x, y) (8) 频率域图像H(x,y) (10)相乘,得到乘积图像(11); (3)对乘积图像(11)进行傅里叶逆变换,得到模板影像与目标影像的相关图像g(x,y)(12),相关图像中的最大值所在的行列值,即模板影像与目标影像达到匹配时的位置;D、交通流参数估算(105): 在车辆提取(103)和车辆匹配(104)的基础上,估算单个车辆行驶速度、路段交通流速度、交通流密度、道路空间占有率、车头距离交通流参数,具体的计算方法如下: (1)单个车辆行驶速度计算: 道路上行驶的单个车辆的行驶速度(V)由下列公式进行计算,单位m/s或km/h: Frequency domain image H (x, y) (10) multiplied by the product of the image (11); (3) the image of the product (11) is an inverse Fourier transform, to obtain the target image and the template image related image g (x , y) (12), where the maximum of the correlation ranks the image values, i.e., the template image and the target image reaches the position matching; D, traffic flow parameter estimation (105): extracting a vehicle and vehicle match (103) ( base 104) on a single estimated vehicle speed, road speed traffic flow, traffic density and road space occupancy, traffic flow parameters from the front, specific calculation method is as follows: (1) a single vehicle traveling speed calculation: travel on the road individual vehicle speed (V) is calculated by the following equation, the unit m / s, or km / h:
Figure CN102855759BC00031
其中,(Xp,yp)与(xm,ym)分别为全色与多光谱影像上对应车辆重心的像素行列值,由步骤车辆提取103和车辆匹配104分别得出;r为影像地面分辨率,单位为m ;t为全色与多光谱拍摄时间间隔,单位为S ; (2)路段交通流速度计算:在连续的自由车流条件下,路段交通流速度用区间平均速度U来表示,单位m/s或km/h: Wherein, (Xp, yp) and (xm, ym) corresponding to the panchromatic and multi-spectral image pixel value of the ranks of the center of gravity of the vehicle, the vehicle extracted by steps 103 and 104 are drawn to match the vehicle; r is ground resolution images, units of m; T is the PAN and multispectral imaging time interval in S; (2) traffic flow velocity calculation: under continuous free traffic conditions, road traffic flow velocity with the average velocity U section represented in m / s, or km / h:
Figure CN102855759BC00032
其中M为路段车辆数量,单位为veh,由车辆提取(103)得到;Vi指第i车观测时的瞬时速度; (3)交通流密度计算: 交通流密度k表示交通流的疏密程度,一定单位长度的道路上所有车辆的数量,单位为veh/km,表不为: Where M is the number of road vehicles, VEH unit, extracted by the vehicle (103) obtained; refers to instantaneous velocity Vi at the i-th observation car; (3) Calculation of traffic density: the density of traffic density k represents the degree of traffic flow, on the road to a certain number per unit length of all vehicles, in units of veh / km, the table is not:
Figure CN102855759BC00033
S为测量道路段的长度,单位为km,根据已有的路网矢量数据获得…为道路段中的车辆数量,单位为veh,由车辆提取(103)得到; (4)交通流量计算: 在已知交通流速度与交通密度的前提下,交通流量q表示为: q = kXu 交通流量(q)等于交通密度(k)与区间平均速度(U)的乘积,单位为veh/h ; (5)道路空间占有率计算: 道路空间占有率(Rs)表示单车道上车辆总长度和路段总长度之比%: S is a measure of the road segment length, in km, is obtained according to the existing road network is VEH ... vector data, obtained by the extraction of the vehicle (103) as the number of vehicles in the road segment units; (4) calculating traffic: in known under the premise traffic flow velocity and traffic density, traffic flow is expressed as q: q = kXu traffic flow (q) is equal to the product of the traffic density (k) and the average interval velocity (U), in units of veh / h; (5 ) road space occupancy calculation: road space occupancy (Rs) represented by the ratio of the total length and total length of the single lane road vehicle%:
Figure CN102855759BC00041
其中S为观测路段长度,根据已有的路网矢量数据获得;Si表示第i辆车的长度,由车辆提取(103)获得; (6)车头距离计算: 车头距离包括空间与时间上的两种形式一车头间距与车头时距,车头间距表示在同向行驶的一列车队中,相邻两辆车的车头之间的距离;车头时距表示在同向行驶的一列车队中,相邻两辆车驶过同一个断面的时间差,路段中所有车头间距的平均值为平均车头间距Hs,单位为m/veli,表示为: Wherein S is the observed length of the link, according to the existing road network obtained vector data; Si represents the i-th length of the car, extracted by the vehicle (103) is obtained; (6) the front distance calculation: distance headway includes two spatial and time when the front forms a pitch headway, headway said in the same direction of travel in a convoy, the distance between the front two cars adjacent; from said in the same direction in front of a train traveling team, two adjacent the car traveled a cross-section of the same time difference, the average of all segment headway average headway Hs, in units of m / veli, expressed as:
Figure CN102855759BC00042
式中:Ksingle为单车道交通流密度,单位是veh/km ;平均车头时距Ht与交通量的关系用如下公式表示,平均车头时距Ht的单位为s/veh: Where: Ksingle lane traffic density is in units of veh / km; Ht and Traffic volume from the average front relationship expressed by the following equation, from Ht units of s / veh average front:
Figure CN102855759BC00043
其中,QsingIe为交通流量,单位为veh/h ; 通过神经网络与面向对象相结合的方法从高分辨率卫星影像中自动提取全色影像道路中行驶的车辆目标;再利用图像频率域相关匹配法,实现全色与多光谱影像中相对应的车辆匹配定位,获得车辆数量、长度以及这两种影像中的车辆重心坐标等信息,最后据此计算出路段乃至区域交通流速度、密度、交通流量、道路空间占有率、车头距离交通流参数,再将这些交通流信息与人工识别或地面监测数据(3)进行对比验证,以检验自动获取数据。 Wherein, QsingIe of traffic units of veh / h; Automatic Extraction of the target vehicle traveling road from the high-resolution panchromatic image by satellite images and object-oriented neural network combination; reuse frequency domain correlation image matching method , with the vehicle to realize a full-color multi-spectral image corresponding to the matching location, obtain information on the vehicle number, and length of the vehicle center of gravity coordinates in the two images, and the last calculated accordingly road traffic speed area, density, traffic , road space occupancy, front, then the traffic flow information identified manually or ground monitoring data (3) from the comparison authentication traffic flow parameters, in order to obtain automatically test data.
CN 201210344373 2012-07-05 2012-09-17 Automatic collecting method of high-resolution satellite remote sensing traffic flow information CN102855759B (en)

Priority Applications (3)

Application Number Priority Date Filing Date Title
CN201210232694 2012-07-05
CN201210232694.2 2012-07-05
CN 201210344373 CN102855759B (en) 2012-07-05 2012-09-17 Automatic collecting method of high-resolution satellite remote sensing traffic flow information

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN 201210344373 CN102855759B (en) 2012-07-05 2012-09-17 Automatic collecting method of high-resolution satellite remote sensing traffic flow information

Publications (2)

Publication Number Publication Date
CN102855759A CN102855759A (en) 2013-01-02
CN102855759B true CN102855759B (en) 2014-04-09

Family

ID=47402307

Family Applications (1)

Application Number Title Priority Date Filing Date
CN 201210344373 CN102855759B (en) 2012-07-05 2012-09-17 Automatic collecting method of high-resolution satellite remote sensing traffic flow information

Country Status (1)

Country Link
CN (1) CN102855759B (en)

Families Citing this family (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103208082B (en) * 2013-03-20 2016-08-10 东南大学 The evaluation of a kind of downtown area road utilization ratio and application process
CN103310199B (en) * 2013-06-17 2016-03-02 武汉大学 High Spatial Resolution Images of the vehicle model identification method
EP3114574A4 (en) 2014-03-03 2018-03-07 Inrix, Inc. Traffic obstruction detection
CN104156622A (en) * 2014-08-27 2014-11-19 北京市环境保护监测中心 Urban traffic particulate matter emission estimation method
CN104794481B (en) * 2015-03-25 2018-01-19 北京师范大学 Merge the sorting technique and device of panchromatic image and multispectral image
CN105513344B (en) * 2015-06-06 2016-08-31 国网山东省电力公司沂水县供电公司 Jam level analysis platforms based on dual communication data
CN104851296B (en) * 2015-06-06 2016-05-04 浙江传媒学院 A kind of section based on remote sensing communication degree detecting system of blocking up
CN104851297B (en) * 2015-06-06 2016-06-29 惠州市妙士酷实业有限公司 A kind of jam level analysis platform based on dual communication data
CN104835329B (en) * 2015-06-06 2016-03-16 福建瑞聚信息技术股份有限公司 Based on Link congestion level sensing detection system in communication
CN104851302B (en) * 2015-06-10 2016-03-16 福建瑞聚信息技术股份有限公司 One kind of network congestion level Identification
CN105023432B (en) * 2015-06-10 2016-08-24 龚麟 Jam level Network Recognition system
CN106097726A (en) * 2016-08-23 2016-11-09 苏州科达科技股份有限公司 The detection determination in region, traffic information detection method and device
CN106897681A (en) * 2017-02-15 2017-06-27 武汉喜恩卓科技有限责任公司 A kind of remote sensing images comparative analysis method and system
CN107862873A (en) * 2017-09-26 2018-03-30 三峡大学 A kind of vehicle count method and device based on relevant matches and state machine
CN108645342A (en) * 2018-04-25 2018-10-12 国交空间信息技术(北京)有限公司 A kind of road width extracting method based on path locus and high resolution image
CN108710875B (en) * 2018-09-11 2019-01-08 湖南鲲鹏智汇无人机技术有限公司 A kind of take photo by plane road vehicle method of counting and device based on deep learning

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6771798B1 (en) * 1998-11-03 2004-08-03 The United States Of America As Represented By The Secretary Of The Navy Hyperspectral visualization extensible workbench
US7058223B2 (en) * 2000-09-14 2006-06-06 Cox Ingemar J Identifying works for initiating a work-based action, such as an action on the internet
CN100418107C (en) * 2006-07-03 2008-09-10 中国科学院遥感应用研究所 Method for fusion generation of high-resolution multi-spectral image
CN101126813B (en) * 2007-09-29 2010-11-24 北京交通大学 High resolution ratio satellite remote-sensing image architecture profile extraction method
US8094960B2 (en) * 2008-07-07 2012-01-10 Harris Corporation Spectral calibration of image pairs using atmospheric characterization
CN101404119A (en) * 2008-11-18 2009-04-08 北京交通大学 Method for detecting and counting urban road vehicle by utilizing remote sensing image
CN101980317A (en) * 2010-09-03 2011-02-23 浙江大学 Method for predicting traffic flow extracted by improved C-V model-based remote sensing image road network

Also Published As

Publication number Publication date
CN102855759A (en) 2013-01-02

Similar Documents

Publication Publication Date Title
Heipke et al. Evaluation of automatic road extraction
Karagiorgou et al. On vehicle tracking data-based road network generation
Kong et al. Vanishing point detection for road detection
Yu et al. 3D reconstruction of road surfaces using an integrated multi-sensory approach
Serna et al. Detection, segmentation and classification of 3D urban objects using mathematical morphology and supervised learning
Jin et al. An integrated system for automatic road mapping from high-resolution multi-spectral satellite imagery by information fusion
Yang et al. Semi-automated extraction and delineation of 3D roads of street scene from mobile laser scanning point clouds
Hinz et al. Automatic extraction of urban road networks from multi-view aerial imagery
Fathi et al. Detecting road intersections from GPS traces
Huang et al. On-board vision system for lane recognition and front-vehicle detection to enhance driver's awareness
CN1984236A (en) Method for collecting characteristics in telecommunication flow information video detection
CN103279759B (en) A kind of vehicle front trafficability analytical procedure based on convolutional neural networks
CN102737510B (en) Real-time traffic condition acquisition method based on mobile intelligent terminal
CN101807352B (en) Method for detecting parking stalls on basis of fuzzy pattern recognition
Sun et al. Vehicle reidentification using multidetector fusion
Zhao et al. Improvement of canny algorithm based on pavement edge detection
CN103077384B (en) A kind of method and system of vehicle-logo location identification
CN104392212B (en) The road information detection and front vehicles recognition methods of a kind of view-based access control model
CN102663394B (en) Method of identifying large and medium-sized objects based on multi-source remote sensing image fusion
Radopoulou et al. Patch detection for pavement assessment
Hu et al. Automatic extraction of main road centerlines from high resolution satellite imagery using hierarchical grouping
CN103500322B (en) Automatic lane line identification method based on low latitude Aerial Images
Maurya et al. Road extraction using k-means clustering and morphological operations
CN103679674B (en) Method and system for splicing images of unmanned aircrafts in real time
CN103500338A (en) Road zebra crossing automatic extraction method based on vehicle-mounted laser scanning point cloud

Legal Events

Date Code Title Description
C06 Publication
C10 Request of examination as to substance
C14 Granted